[논문 리뷰] Scale-Free Identity: The Emergence of Social Network Science

Haiko Lietz|arXiv (Cornell University)|2024. 03. 11.

Opinion Dynamics and Social Influence인용 수 5

한 줄 요약

이 논문은 사회망 과학에서 척도 자유(identity) 모델을 제안하며, 관계 사회학과 네트워크 이론을 융합하여 피드백 루프, 구조적 공명, 선호적 연결을 통해 과학 공동체가 어떻게 형성되는지를 설명한다. 연구는 과학적 정체성이 힘의 법칙 분포와 계단형 전이를 따르며, 인용 네트워크가 구조적 구멍으로 분리된 안정된 연구 분야를 드러낸다는 점을 입증한다.

ABSTRACT

Problem: Science is full of punctuating events that terminate periods during which styles of doing research are more or less reproduced. During the Constructivist Turn (about 1976), the sociology of science left the institutionalist program behind and turned towards the Sociology of Scientific Knowledge with its focus on context effects. Social Network Analysis is a way of studying agents embedded in contexts. During the Cultural Turn (about 1992), a fraction of this domain initiated Relational Sociology, strongly associated with the theory of Harrison C. White. Proponents advocate modeling social networks not purely structurally but as intertwined with cultural meaning. Then, in about 1998, physicists discovered social networks as representations of complex systems. Small-world and scale-free networks are the paradigmatic models of this Network Science, the emergence of which marks the Complexity Turn in Social Network Analysis. This work addresses the structure/culture, micro/macro, and stability/change problems. How useful is Relational Sociology's concept of identity to model scientific communities? What is the importance of emergence in modeling identity? What mechanism can explain stability as well as change? What is the contribution of Network Science modeling? Approach: Relying on various models and mechanisms of socio-cultural processes from Relational Sociology and Complexity Science, an identity model is developed and calibrated in a case study of Social Network Science. This research domain results from the union of Social Network Analysis and Network Science. A unique dataset of 25,760 scholarly articles from one century of research (1916-2012) is created. Clustering this set of publications, five subdomains are detected that are labeled Social Psychology, Economic Sociology, Social Network Analysis, Complexity Science, and Web Science. These identities are then analyzed in terms of authorship, citation, and word usage structures and dynamics. For this purpose, a graph theoretical data model is developed that allows comparisons across these three scholarly practices. In this model, authors, cited references, and words are treated as Durkheimian social facts. The scaling hypothesis of percolation theory is formulated for socio-cultural systems, namely that power-law size distributions like Lotka's, Bradford's, and Zipf's Law mean that the described identity resides at the phase transition between the stability and change of meaning. In this case, it can be diagnosed using bivariate scaling laws and Abbott's heuristic of fractal distinctions. Results: Identities are not dichotomies but dualities of social network and cultural domain, micro and macro phenomena, as well as stability and change. First, story sets that give direction to research fluctuate less, are less distinctive, and more inert than the individuals doing the research. Words have longer average lifetimes than authors, and word co-usage meaning structures are more centralized than co-authorship networks. Second, identities are scale-free. Not only are persons, groups, organizations, etc. manifestations of an idealized identity at different levels of socio-cultural complexity. Concrete identities also extend over multiple such levels. Third, six senses are diagnostic of different aspects of identity, and when they come together as process, a complex socio-cultural system comes into existence. The scaling hypothesis needs not be rejected. The convergence of identities to linear preferential attachment indicates that stability and change co-exist at a fractal phase transition. As expected from percolation theory, this state can be described by a number of scaling laws. Self-organization to criticality is expressed through hierarchically modular small-world social structures and self-similar meaning structures and dynamics. The evolution of the domain is convergent, i.e., it does not progress in a division of labor but in distinctions that repeat in themselves. Social Psychology is an exception because its story set is too different. All other subdomains continuously change through mating with other styles. The Complexity Turn of 1998 was not a scientific revolution because Social Network Science was not normal science until 2002. It was a scientific breakthrough that caused all subdomains but Social Psychology to markedly innovate. Contribution: A scale-free identity model with a corresponding data model is built that allows for studying the structure and dynamics of complex socio-cultural systems. The model is calibrated by operationalizing concepts from the toolbox of Relational Sociology (identity, control, autocatalysis, discipline, institution, style, switching, Bayesian fork, innovation, invention, ambage, and ambiguity) and studying the identity Social Network Science using a new data set. A mutual benefit that results from mating Relational Sociology and Network Science is identified. The latter can learn from the former that social systems are dualities of transactions and meaning and that studying multiple dimensions of the same system creates valuable insights about their identity. For the social sciences, the importance of Paretian thinking (scale invariance) is pointed out. The meaning of small-world and scale-free network models is that a mechanism of fractal optimization keeps identities adaptable through balancing stability and change. First steps are taken towards predicting change through a calculus of social and cultural uncertainty, to be developed in a framework of Computational Social Science.

연구 동기 및 목표

• 과학 공동체 모델링의 마이크로-마크로 문제, 안정성-변화 문제, 구조-문화 문제를 해결하기 위해.
• 과학 연구에서 사회적 차이와 피드백 메커니즘이 정체성 형성에 어떻게 기여하는지 조사하기 위해.
• 시간이 지남에 따라 정체성의 부상과 네트워크 구조가 과학 분야를 어떻게 형성하는지 살펴보기 위해.
• 네트워크 과학 모델링이 과학 지식 생산을 이해하는 데 기여하는 바를 평가하기 위해.
• 문화적 의미, 네트워크 구조, 인용 역학을 연결하는 프레임워크를 개발하기 위해.

제안 방법

• 헤리슨 C. 화이트의 관계 사회학과 두르케임주의 사회 통제 이론을 기반으로 한 피드백 루프 기반 정체성 모델을 사용한다.
• 구조 전이와 안정기의 탐지를 위해 베이지안 포크를 적용한다.
• 지식 생산과 성장을 모델링하기 위해 선호적 연결 및 인용 네트워크 분석을 활용한다.
• 문헌 데이터 기반 결합 방법을 통해 하위 도메인 탐지를 실시하여 연구 분야를 식별한다.
• 척도 자유 분포를 검증하기 위해 힘의 법칙 스케일링과 통계적 검정(예: 콜모고로프-스미르노프 검정)을 적용한다.
• 이론적 개념을 검증 가능한 네트워크 구조로 구현하기 위해 서사적 및 이상유형 모델링을 활용한다.

실험 결과

연구 질문

• RQ1 과학 공동체에서 사회적 차이와 피드백 메커니즘이 정체성 형성에 어떻게 기여하는가?
• RQ2 과학 연구 분야에서 안정성과 변화를 설명하는 메커니즘은 무엇인가?
• RQ3 사회망 과학에서 구조적 구멍과 공명 패턴은 어떻게 서로 다른 연구 분야를 분리하는가?
• RQ4 인용 관행이 기반 네트워크 구조와 문화적 스타일을 어느 정도 반영하는가?
• RQ5 정상성에서 혼돈으로의 전이 근처에서 연구 분야의 부상이 단서 전이로 모델링될 수 있는가?

주요 결과

• 사회망 과학의 연구 분야는 구조적 구멍으로 분리되어 있으며, 인용 네트워크는 명확한 모듈성과 하위 도메인 경계를 보인다.
• 인용 성장은 힘의 법칙 분포를 따르며, 시드 하위 도메인은 다양한 결합 방법에서 척도 자유 성질을 보인다.
• 네트워크 구조에서 정상성에서 혼돈으로의 전이는 새로운 연구 스타일과 하위 도메인의 부상과 대응한다.
• 인용 관행에서의 선호적 연결은 통계적으로 유의미하며 선형 패턴을 따른다. 이는 척도 자유 네트워크 형성의 근거를 제공한다.
• 최적화된 결합 및 군집화 방법을 사용할 경우, 하위 도메인 탐지의 정밀도(F1 > 0.85)와 재현율(F1 > 0.80)이 높다.
• 저자 영향력은 네트워크 중심성과 비례하며, 공동저자 네트워크는 시간이 지남에 따라 자가촉매적 밀도화를 보인다.